Reading for Today's Lecture:
Goals of Today's Lecture:
Today's notes
In any model the statistic
is sufficient. In any iid
model the vector of order statistics
is
sufficient. In the
model then we have three possible sufficient
statistics:
Notice that I can calculate S3 from the values of S1 or S2but not vice versa and that I can calculate S2 from S1 but not vice
versa. It turns out that
is a minimal sufficient
statistic meaning that it is a function of any other sufficient statistic.
(You can't collapse the data set any more without losing information about
.)
To recognize minimal sufficient statistics you look at the likelihood function:
Fact: If you fix some particular
then
the log likelihood ratio function
The
subtraction of
gets rid of those irrelevant constants
in the log-likelihood. For instance in the
example we
have
FACT: A complete sufficient statistic is also minimal sufficient.
Hypothesis testing is a statistical problem where you must choose,
on the basis of data X, between two alternatives. We formalize this
as the problem of choosing between two hypotheses:
or
where
and
are a partition of the model
.
That is
and
.
A rule for making the required choice can be described in two ways:
For technical reasons which will come up soon I prefer to use the
second description. However, each
corresponds to a
unique rejection region
.
The Neyman Pearson approach to hypothesis testing which we consider first treats the two hypotheses asymmetrically. The hypothesis Ho is referred to as the null hypothesis (because traditionally it has been the hypothesis that some treatment has no effect).
Definition: The power function of a test (or the corresponding critical region
)
is
We are interested here in optimality theory, that is,
the problem of finding the best .
A good
will
evidently have
small for
and large for
.
There is generally a
trade off which can be made in many ways, however.
Finding a best test is easiest when the hypotheses are very precise.
Definition: A hypothesis Hi is simple
if
contains only a single value
.
The simple versus simple testing problem arises when we test
against
so that
has only two points in it. This problem is of importance as
a technical tool, not because it is a realistic situation.
Suppose that the model specifies that if
then
the density of X is f0(x) and if
then
the density of X is f1(x). How should we choose
?
To answer the question we begin by studying the problem of
minimizing the total error probability.
We define a Type I error as the error made when
but we choose H1, that is,
.
The other kind of error, when
but we choose
H0 is called a Type II error. We define the level
of a simple versus simple test to be
Suppose we want to minimize
,
the total error
probability. We want to minimize
Theorem: For each fixed
the
quantity
is minimized by any
which has
Neyman and Pearson suggested that in practice the
two kinds of errors might well have unequal consequences.
They suggested that rather than minimize any quantity
of the form above you pick the more serious kind of error, label it Type I and require your rule to hold the
probability
of a Type I error to be no more than
some prespecified level
.
(This value
is typically 0.05 these days, chiefly for historical reasons.)
The Neyman and Pearson approach is then to minimize beta subject to the constraint
.
Usually this is really equivalent to the constraint
(because if you use
you could make R larger and keep
but
make
smaller. For discrete models, however, this
may not be possible.
Example: Suppose X is Binomial(n,p) and
either p=p0=1/2 or p=p1=3/4. If R is any critical region
(so R is a subset of
)
then
The problem in the example is one of discreteness. Here's how we get
around the problem. First we expand the set of possible values of to include numbers between 0 and 1. Values of
between
0 and 1 represent the chance that we choose H1 given that we observe
x; the idea is that we actually toss a (biased) coin to decide!
This tactic will show us the kinds of rejection regions which are
sensible. In practice we then restrict our attention to levels
for which the best
is always either 0 or 1. In the
binomial example we will insist that the value of
be either 0 or
or
or ...
Definition: A hypothesis test is a function whose values are always in [0,1]. If we observe X=x then we
choose H1 with conditional probability
.
In this case we have
Theorem: In testing f0 against f1 the probability
of a type II error is minimized, subject to
by the test function:
Example: In the Binomial(n,p) with p0=1/2 and p1=3/4the ratio f1/f0 is
NOTE: No-one ever uses this procedure. Instead the value of used in discrete problems is chosen to be a possible value of the rejection
probability when
(or
). When the sample size is large
you can come very close to any desired
with a non-randomized test.
If
then we can either take
to be 343/32
and
or
and
.
However, our
definition of
in the theorem makes
and
.
When the theorem is used for continuous distributions it can be the case
that the cdf of
f1(X)/f0(X) has a flat spot where it is equal to
.
This is the point of the word ``largest'' in the theorem.
Example: If
are iid
and
we have
and
then
The rejection region looks complicated: reject if a complicated
statistic is larger than
which has a complicated formula.
But in calculating
we re-expressed the rejection
region in terms of
Definition: In the general problem of testing
against
the level of a test function
is
Application of the NP lemma: In the
model
consider
and
or
.
The UMP level
test of
against
is
Proof: For either choice of
this test has level
because
for
we have
(Notice the use of .
The central point is that the critical
point is determined by the behaviour on the edge of the null hypothesis.)
Now if
is any other level
test then we have